Comparing Collocations Across Languages:
An English-Czech Sample
Aleš Klégr 1
1. Contrastive Comparison of Collocations
The importance of the concept of collocation, a recurrent combination of words, has
long been recognized in theoretical linguistics. However, its relevance for the applied
sphere, language teaching, lexicography and translation (including machine
translation), etc., is just as obvious. The concept has caught the imagination of
professionals in applied linguistics and ELT and is currently getting more and more
attention. This is documented by textbooks such as McCarthy and O’Dell’s English
Collocations in Use (2005) and others. The interest is understandable inasmuch as
collocations are recognized as a source of naturalness in speech, and naturalness is
one of the primary goals in language teaching. The inevitable limitation of English
textbooks for foreign learners and even of the best learner’s dictionaries is that
without knowing how collocations in English relate to those of the learner’s native
language, they cannot alert the learner to the pitfalls of interference and asymmetries.
Obviously contrastive analysis of collocations is called for, but research in this
direction is still somewhat neglected.
There are several difficulties associated with collocations. For one thing, there
is still a lack of consensus on their definition. Partington (1998) divides the definitions
of collocations into textual (co-occurrence in a text), statistical (co-occurrence with
greater than random probability) and psychological (co-occurrence due to a
psychological link between words). Hoey (2005), discussing the merits of each type,
opts for a statistical and psychological approach. It is not clear, however, whether a
collocation should be identified with a syntagma or not. Collocation as a
psychological association argues for a syntagmatic nature, yet from a statistical view
this is an unnecessary restriction. In the following we shall take the position that a
prototypical collocation is syntagmatic. Another difficulty with collocations is the
absence of clear criteria that would delimit the range of a node’s collocates and
provide a firm basis for their (lexicographic) description. The problem is encountered
in monolingual dictionaries, but in collocational dictionaries it becomes crucial.
Nuccorini (2003), who analyzes several collocational dictionaries in terms of their
features and congruence between the professed methodology and aims and the success
of their implementation, concludes that these dictionaries are extremely useful for
advanced users but finds that the definition and classification of collocations, together
with terminological confusion, constitute a major stumbling-block and inevitably
affect the criteria for the selection and treatment of both headwords and their
collocates. The following is an attempt to address some of these issues.
1
Department of English, Charles University, Prague
e-mail: [email protected]
2. Comparing Collocations in Bilingual Corpora
Cross-language comparison of collocations drawing on bilingual corpora can make
use of two types of corpus, comparable and parallel. Each type offers a somewhat
different kind of information on collocations, provides different possibilities and
involves different snags. With two independent comparable corpora, neither language
is influenced by the other and the set of collocations found for a given node thus
reflects a situation specific to either of the languages, whether it is the range or variety
of collocations occurring in the environment of the node or their overall number. A
parallel corpus, on the other hand, shows which and how many collocations of the SL
translate compositionally in the TL, and which of them represent a complex lexical
choice requiring a special translation solution. The main focus of this paper is a
comparison based on comparable Czech and English corpora and describing two
equivalent nodes by means of a simple, but as we hope effective strategy. In the final
part, a few remarks on comparing collocations in a parallel corpus are made.
3. Comparing Collocations in Comparable Corpora:
The Case of sadness/smutek
Experience from compiling a Czech-English noun-verb combinatory dictionary
(Klégr et al., 2005) shows that even with one type of syntagma a noun will typically
combine with hundreds of verbs, most of which occur only once. This poses the
problem of collocate selection and presentation. To give a full range of a word’s
collocates found in a corpus is too much space-consuming, and is likely to prove
bewildering for the dictionary user. To include only habitual collocations such as
heavy smoker, rancid butter or statistically significant collocations (cf. Sinclair et al.,
2004) would deprive the user of stylistic choices and fail to provide an adequate
picture of the word’s collocational potential. Similarly, to include collocates of a
given node “down to a frequency cut-off point [the top 20 collocates], thereby
automatically giving due weight to the most frequent cases” (Stubbs, 2002) has its
drawbacks as the most frequent collocates prove to be the most general and least
informative ones (cf. also Sinclair’s upward collocates, 1991).
3.1 The Sample and Methodology
The study makes use of the fact that for each language there is a corpus namely the
British National Corpus (2nd ed.) and Syn2000 compiled by the Czech National
Corpus Institute. They were made available in the same year (2000) and are
comparable in size (100 million words), methodology, and the range of texts. The
choice of sadness/smutek as the node was motivated by the fact that an abstract noun
tends to be monosemic, has simple morphology, and relatively straightforward
translation. Still, the corpus data showed one striking difference between them:
sadness had 751 (relevant) hit lines the BNC, smutek 2327 hit lines in Syn2000. The
discrepancy is probably due to three factors: (i) polysemy of the Czech smutek; (ii)
sadness having several widely used synonyms as competitiors, while the synonyms
of smutek are formal and much more restricted in use; (iii) sadness being a complex
word, unlike the synchronically simplex smutek (the difference in valency behaviour
between derived and underived words is mentioned by Čermák, 2005). The analysis
2
focused on the V-O type of collocation. Although the noun functions as a
complement, the lexicographic experience and research in natural language
generation suggest that the noun is indeed the starting point. As Heid (1994) notes,
“for noun-verb-collocations (e.g. in the verb-object case: the object noun must be
determined first, only then a collocationally adequate verb can be selected)”. Despite
the efforts to choose nodes with limited collocability and the restriction to the
syntactically and semantically relatively clean-cut V-O syntagma (with the node as
object and the collocate as finite verb), the number of collocates in the corpora and
the task of identifying grammatically relevant collocations still proved daunting.
The first step in the analysis was to prepare a complete list of collocates for
each node of a given syntactic type (V-O) from the total of all collocates obtaining in
each corpus, taking care that in either language the same lexical unit (sense) is
analyzed. The resultant sets of collocates were then subject to semantic analysis. The
central premise was that while the number of collocates of the node may amount to
hundreds, the semantic range of these collocates is much smaller. Accordingly the
verbal collocates of sadness/smutek were divided into subsets of broadly substitutable
items called synsets (i.e., forming sets composed of synonyms, hyponyms,
hyperonyms; negative verbs forms – and their lexical parallels (antonyms) – are
subsumed under the positive forms, the presence of an antonym in the subset is
marked by (not)). Each synset is considered to represent one type of broadly
conceived, but distinct node-collocate relationship. Their sum is expected to
distinguish one node from another. This procedure reduced the total sets of verbal
collocates of sadness/smutek to a relatively small number of synsets, 19 and 30
respectively, describing the collocational preferences of each node. The number and
type of the resultant synsets (and their frequencies) then made it possible to compare
the similarities and dissimilarities of the collocational sets of sadness/smutek both
quantitatively and qualitatively. Each collocational synset can be described by two
features: the number of distinct lexical items (types) it includes (lexical variety), and
the total of occurrences of the verbs in the synset (token frequency). Needless to say,
the semantic classification of the collocates is rather tentative. The collocate synsets,
marked by square brackets, are usually designated by the most frequent (prototypical)
verb within each subset (or its English translation in the case of Czech synsets).
Semantically isolated collocates form separate synsets and their own headings.
3.2 The Verbal Collocates of sadness
The node sadness was found to be a finite-verb object in 205 collocations. It is not
without interest that only 23, i.e. 26.4 percent of the verbal collocates have a
frequency higher than 2. In other words, almost 3/4 of the verbs co-occurring with
sadness appear just once. The respective instances (tokens) of the finite-verb
collocates were assigned to 87 types (lemmas). The lemmas, in turn, were divided
into 19 synsets (Table 1). We take it that these 19 synsets exhaustively summarize all
activities performed with sadness (as object) in the BNC.
The first six largest synsets among the verbs collocating with sadness as
object are [feel], [express], [show (not)], [cause], [perceive] and [deal with].
Although they represent only 1/3 of the semantic synsets found with this noun, they
account for 76.2 percent of its collocate tokens. In other words, speaking of sadness
English speakers most frequently say that they feel, express, show/hide, speak of its
cause, perceive or deal with it (by avoiding, controlling it, etc.). Among the six,
3
[feel] alone is responsible for 31.2 percent of tokens, which is in fact more than twice
the percentage of the second largest synset [express]. Thus [feel], the richest synset
in types and tokens alike, includes 13 verbs (feel 41x, have 6x, sense 4x, etc.) which
account for 64 of the V-O collocations with sadness. These verbs express various
kinds of feeling. While most of them can be regarded as (partial or near) synonyms
(feel, sense, have, experience), the last four are somewhat different: be drawn to,
tend towards sadness describe incipient feeling, allow os is a condensation of allow
os to feel, and love describes how we feel about the feeling.
Each of the 19 synsets appears to have one principal verb realizing the synset
which accounts for the type/token asymmetry. The poorer the lexical variety, the
higher the correspondence between the number of types and tokens. While the token
frequency tells us whether sadness is often or, conversely, hardly ever thought of in
terms of the activity of the given semantic synset, the lexical variety presumably
reflects the different possibilities of realization or stylistic variation (feel/harbour
sadness).
1.
synset
heading
feel
number
of types
13
2.
express
11
3.
show (not )
9
4.
cause
8
5.
6.
perceive
deal with
4
11
7.
8.
9.
10.
11.
know
include
share
relieve (not)
remember
(not)
think of
associate
with
respond to
change to
explain
relate to
attribute
find
Total:
12.
13.
14.
15.
16.
17.
18.
19.
verbs (types) in the synset
2
4
1
5
2
feel (41x) , have (6x), sense (4x) , experience (3x), endure
(2x), bear, harbour, realize, suffer, love, be drawn to, tend
towards, allow os
express (14x), speak of (5x), convey, talk about, tell of ,
write of, portray, capture, sum up, measure, mean
show (6x), reveal, indicate, display, bring closer; hide (4x),
conceal (2x), disguise (2x), deny
bring (7x), cause (5x), move to (2x), bring up, evoke,
induce, provoke, work up
see (12x), glimpse, hear of, watch
avoid, control, counteract, dispel, heal (sb) of, outweigh,
put aside, stop, ward off, fill up, sort out
know (about, of)(6x), learn about
include (2x), hold (2x), contain (2x), carry (with it)
share (7x)
ease (2x), release, release from, relieve, deepen
remember (3x), forget
2
3
3
3
2
2
1
1
87
number of
tokens %
64
31.2
28
13.7
19
9.3
19
9.3
15
11
7.3
5.4
7
7
7
6
4
3.4
3.4
3.4
2.9
1.9
think of (2x), reflect on
associate (noise) with, mingle with, mix with
3
3
1.5
1.5
respond to, acknowledge, accept
change (from) to s., cool into, give way to
explain, account for
be (all) about, be concerned with
attribute (to st)
find
3
3
2
2
1
1
205
1.5
1.5
1.0
1.0
0.5
0.5
100
Table 1: Semantic synsets of the verbal collocates of sadness (in V-O).
3.3 The Verbal Collocates of smutek
The number of the Czech verbal collocates participating in the V-O syntagma with
smutek is, on the whole, proportionate to the higher incidence of the word smutek in
the Czech corpus: the search yielded 235 distinct verbs (types) with 453 tokens.
Counting verbs in Czech texts is somewhat more complicated than in English due to
4
greater form variation (aspectual variants, negative and reflexive forms, etc.).
Aspectual variants were generally disregarded where the variants differed only in
terms of perfectiveness, repetition, etc. (cítit/pocítit, vyjádřit/vyjadřovat,
skrýt/skrývat). Synthetic negative verb forms were subsumed under positive forms
(just as the English (not) remember/forget were placed in the same synset). Reflexive
forms with se are marked as such – zbavit (se) – but not counted separately. The case
and number variability of the Czech node smutek (a plural form, for instance,
appeared in almost a hundred cases) was disregarded.
As might be expected, the larger number of collocates in Czech produced a
larger number of semantic synsets: 30 in all instead of the 19 in English. Perhaps
surprisingly, the first six largest synsets are the same ones as for sadness, [feel], [deal
with], [cause], [express], [show (not)], and [perceive]. However, with the exception
of [feel], which is the most frequent in either language, the order of the categories is
markedly different. The most conspicuous synset in Czech is [deal with]. Although
second in frequency after [feel], the difference is quite small: [feel] 23.8 percent,
[deal with] 19.9 percent. It certainly surpasses the [feel] synset in lexical diversity:
while [feel] is realized by 31 verbs (13.2 percent), including such special ‘feel’
words as vychutnávat-relish, libovat si v-delight in, etc., [deal with] includes 63
distinct verbs (26.8 percent of the total), which is more than twice as many. The most
frequent [deal with] verb, rozmlouvat-talk out of (11x) has no counterpart in English.
By contrast, there is no counterpart in the Czech synset to the English control or
avoid sadness. It is difficult to say how much can be read into finings. It seems that
Czech prefers to ‘deal with sadness’ in a more varied way and that, whatever the
manner, ‘dealing with’ sadness is almost as popular, if not more, as ‘feeling’
sadness. The question remains which of the two, token frequency or lexical variety,
is to be given more weight.
synset heading
1.
feel
no.
of types
31
2.
deal with
63
3.
cause
21
verbs (types) in the synset
cítit (27x), mít (24x including 8x: na/v očích, v sobě, v tváři,
na duši), pocítit (11x), prožívat (7x), pociťovat (6x), prožít
(3x), necítit (3x), zažít (3x), trpět 7 (2x), dostat, nasávat,
nezažívat, nosit v sobě, unést, vléci s sebou, procítit,
rozmlouvat se 7, soucítit se 7, utápět se v 6, vydržet,
zakoušet, zakusit, zažívat // bát se 2, libovat si ve 6,
postrádat, radovat se ze 2, stydět se za, tíhnout ke 3, velebit,
vychutnávat (něčí)
rozmlouvat (11x), zahánět (4x), zaplašit (4x), přehlušit (3x),
vyrovnat se se 7 (3x), odložit (2x), odnášet (2x), překonat
(2x), řešit pl (jídlem) (2x), utápět (v sobě) (2x), zapíjet (2x),
zpracovávat (2x), odpustit, odvrátit, prominout, uzamknout,
zapít, bojovat proti 3 (prací), čelit, dávkovat, chránit proti 3,
nebránit se 3, nedbat 2, nechat doma, nechávat za sebou,
nepřemáhat, obelstít, odbýt si, odehnat, odnést, odplavit,
pohřbít, pomoci od 2, popřít, probudit ze 2, překonávat,
překousnout, přemoci, přepíjet, rozkoukat se ze 2, setřást ze
sebe, shodit, sloužit proti 3, spláchnout, stírat, ubránit se 3,
utopit 4pl, vydýchávat 4pl, vykoupit, vymanit se ze 2,
vymazat, vyplakat 4pl, vyrovnávat se se 7, vysvobodit z 2,
vyzrát na, zabránit 3, zahnat, zahodit 4pl, zapudit, zaříkávat ,
zbavit 2, zbavit se 2, odstranit
vyvolávat (10x), přinést (7x), způsobit (5x), vyvolat (4x),
přinášet (3x), uvrhnout do 2 (3x), budit (2x), evokovat (2x),
(vrána) nosit (2x) , vzbudit (2x), rozprostírat, vdechnout
(pohledu), vystavit 3, kouzlit (= vykouzlit), nechat to v kom,
probouzet, probudit, vhánět, vnášet, vykouzlit, vzbuzovat
number of
tokens %
108 23.8
90
19.9
51
11.3
5
4.
express
17
5.
show (not)
22
6.
perceive
15
7.
know
11
8.
change to
9
relieve
(not)
10. respond to
11. pretend
12. associate with
8
4
4
4
13. remember (not)
14. share
15. include
2
1
4
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
3
2
1
2
1
1
1
1
1
1
1
1
1
1
1
235
9.
disturb
compensate
relate to
use
resemble
caress
expect
get used to
return to
participate in
continue
lose
mistake for
need
originate from
Total:
vyjadřovat (11x), vyjádřit (10x), znamenat (7x), sdělovat
(3x), dát průchod 3, hučet si (svůj), uvádět, hovořit o 6,
malovat, mluvit o 6, nevypovídat o 6, přiblížit, říci,
tématizovat, vykřičet, vyprávět o 6, neznamenat
neskrývat (6x), skrýt (4x), skrývat (4x), vyzařovat (3x),
ukázat (2x), tajit (2x), zakrývat (2x), dát najevo (2x), odhalit
(2x), projevovat (2x), dát na sobě znát , nedávat na sobě znát,
nejevit, neprojevit, netajit, netajit se 7, nezastírat, ukazovat,
vyjevit, vyjevovat, ukrývat, zastírat
vidět (10x), spatřit (na tváři, v očích) (3x), ohlédnouti se po 6
(2x), setkat se se 7 (2x), číst s. (v obličeji, z očí) (2x), uvidět
(2x), vycítit (2x), představit si, nevidět, přihlížet 3, uslyšet,
všimnout si 2, vyčíst (z pozdravení), zahlédnout,
zaznamenávat
pochopit (3x), poznat (3x), znát (3x), chápat (2x), nechápat,
neznat, porozumět 3, reflektovat, rozeznat, rozpoznat, vědět o
6
propadnout 3 (3x), střídat (radost) 4 (2x), propadat 3, dojít
k 3, neupadnout do 2, propadnout se do 2, upadnout do 2,
změnit se v, zvrhnout se do 2
násobit (2x), přidat, neakcentovat, podtrhovat, tišit, umocnit,
uvolnit , ztlumit
uklonit se 3 (3x), přijmout (2x), respektovat (2x), reagovat na
nehrát (2x), hrát, předstírat, přehrávat
vměšovat s. (do čeho), mísit se se 7, přidružit se k 3, spojovat
(X,Y) s.
zapomenout na, zapomínat na
sdílet
zahrnovat, nepočítat (mezi co), vložit (do hudby), vzít (s
sebou)
poskvrnit, rušit v 6, vyrušit ve 6
vynahradit, vyvažovat (radostí)
týkat se 2
využít 2, přisoudit 3 (roli)
podobat se 3
pohladit
očekávat
zvyknout si na
vrátit se do 2
podílet se na 6
setrvávat ve 6
ztrácet
zaměňovat se 7
potřebovat
vzniknout ze 2
44
9.8
41
9.1
31
6.9
18
4.0
12
2.7
9
2.0
8 1.8
5 1.1
4 0.9
4 0.9
4 0.9
4 0.9
3 0.7
2 0.4
2 0.4
2 0.4
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
1 0.2
453 100
Table 2: Semantic synsets of the verbal collocates of smutek (in V-O).
Note on column 4: numbers in round brackets give frequency higher than 1;
unbracketed numbers show case of the object
As with sadness, there seem to be few, if any collocations of smutek that
would display mutual or at least unidirectional expectation on the part of the noun or
the verb. A possible candidate is zapít or utopit (‘drown’), though in Czech as in
English žal sorrow(s) rather than smutek/sadness is the first choice. In common with
English, cítit-feel is the most frequent verb, and obviously the number-one collocate,
though it can hardly be called sadness-specific. On the whole, there is an even
greater proportion of one-off instances than in English. Although they sound
perfectly natural, they are highly context-specific. Only few of them are distinctly
metaphoric (pohladit-caress, poskvrnit-tarnish/sully).
6
3.4 Contrastive Analysis of the sadnes/smutek Collocates
Some of the differences have already been mentioned above: the V-O collocations
differ in absolute figures, which is due to different representations of sadness and
smutek in the respective corpora. They differ in the number and type of semantic
synsets they have. In the total of 34 synsets identified in the verbal collocates of
sadness/smutek, 15 are common to both sadness/smutek, 19 are specific:
synsets shared
synsets only in English
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
1.
2.
3.
4.
cause
deal with
express
feel
perceive
show (not)
change to
include
know
relieve (not)
remember (not)
respond to
share
associate with
relate to
attribute
explain
find
think of
synsets only in Czech
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
resemble
caress
compensate
continue
disturb
expect
get used to
lose
mistake for
need
originate from
participate in
pretend
return to
use
Table 3: Comparison of the distribution of sadness/smutek synsets in English and
Czech.
In terms of proportional representation the synsets can be conveniently, though
somewhat arbitrarily, divided into three zones: the core group with token frequencies
of 5 percent and more (in bold type in Table 3), a mid-level group of 5–2 percent (in
italics), and a marginal group below 2 percent (normal type). The fact that
sadness/smutek share fifteen, i.e. almost half, of the synsets distinguished, especially
the core and mid-level ones, is a highly significant finding as it suggests that the
collocational range of the two nodes is very similar. This is accentuated by another,
probably even more important finding that the core group in either language consists
of the same six synsets which account for 76.2 percent of collocations in the English
sample and 80.8 percent in the Czech one. The six synsets can be seen very much as
defining the collocational preferences of sadness/smutek as regards their transitive
verb collocates.
The 21 marginal synsets with frequencies below 2 percent, on the other hand,
signal the potential, rather than normal collocations of sadness/smutek. Two things
are worth noticing. Firstly, only two of them are shared, [associate with] and [relate
to]. The lack of overlap between the English and Czech marginal synsets may partly
be a function of their low incidence; partly it may signify different frames of thought
in either language. The English speakers apparently do not think of sadness in the
same terms or contexts as the Czech speakers, although it would be easy to match the
Czech verbs with English ones and vice versa. The other remarkable thing is the
semantic diversity of the Czech marginal synsets. In fact, they are as numerous as the
synsets shared by both nodes. This could be attributed to the higher incidence of
smutek in the Czech corpus, but it may also be the case that the positions of sadness
7
and smutek in their respective lexical fields are simply different. We may
hypothesize that sadness is less important compared to smutek, being frequently
replaced by synonyms such as unhappiness, sorrow, grief, etc. As a consequence, it
appears in fewer contexts than its Czech counterpart.
The core groups of the synsets (see Table 4) deserve closer attention. The fact
that the first six are the same for both languages suggests similar collocational
preferences of sadness/smutek and, by the same token, similar communicative
preferences in the two speech communities. Evidently, for speakers of English and
Czech conveying that they ‘feel sadness’ has top priority (31.2 percent and 23.8
percent respectively). Conversely, the finding that these synsets are differently
distributed and show different lexical varieties suggests subtle dissimilarities between
the communicative preferences in either speech community. While in English the
second most frequent synset [express] has less than half the percentage of [feel] and
the last of the six, [deal with], is actually very near the cut-off point of 5 percent, in
Czech [deal with] comes second with a proportion quite close to [feel] and [express]
is sandwiched between [cause] and [show (not)] in the middle. The position of [cause]
and [show (not)] is switched in English, though again roughly in the middle. The
representation of [perceive] is about the same in both languages. We may conclude
that, quite interestingly, speakers of English primarily seem to feel and express
sadness, whereas speakers of Czech feel and deal with it.
sadness collocates
semantic synset
types
%
tokens
feel
13
14.9 64
31.
2
express
11
12.6 28
13.
7
show (not )
9
10.3 19
9.3
cause
8
9.2
19
9.3
perceive
4
4.6
15
7.3
deal with
11
12.6 11
5.4
%
smutek collocates
semantic synset types
%
tokens
feel
31
13.2 108 23.8
deal with
63
26.8 90
19.9
cause
21
8.9
51
11.3
express
17
7.2
44
9.8
show (not)
22
9.4
41
9.1
perceive
15
6.4
31
6.9
%
Table 4: Comparison of the six core synsets of sadness/smutek.
When lexical diversity is taken into account, the order of synsets in English
remains the same, except for [deal with] which draws level with [express]. In Czech
there are two marked changes compared to token frequency: [deal with] is by far the
most varied lexically, in fact twice as much as [feel], and [show (not)] comes third
before [cause]. In sum, lexical variety somewhat raises the role of [deal with] in
English but generally is in agreement with token frequency, whereas in Czech it
underscores the importance of [deal with] quite dramatically compared to [feel] and
accentuates the role of [show (not)]. The overall picture of sadness/smutek
collocations thus speaks of similar speech habits but distinct specific tendencies and
preferences.
The strategy of describing collocations through their synsets appears capable
of handling the collocability of a node in an economical way by reducing the
hundreds of collocates to a few groups. Fears that the large number of one-off
collocates might result in too many synsets proved unnecessary. The method works
equally well in one language or in two, making possible a meaningful contrastive
8
analysis of equivalent nodes. By assigning their numerous collocates to synsets it
manages to arrive at a coherent picture of the collocational preferences of the
contrasted nodes, highlighting their similarities and dissimilarities. To some extent, it
may also be predictive of potential collocates (synonyms, co/hyponyms, etc., of the
actual collocates).
4. Comparing collocations in a parallel corpus
Collocations in a parallel corpus involve SL collocations and their translation
equivalents in the TL which may, but need not be structurally parallel (cf. McDonald
et al., 2004). While SL collocations reflect authentic collocational patterns of the
source language, their TL equivalents depend on the degree of isomorphism between
the two languages. In principle, the translation will be a continuum running from a
node-collocate match (direct or compositional translation) to the complete absence of
an equivalent (non-match or lexical gap). In between are various forms of indirect
translation whereby equivalence is achieved at the expense of formal correspondence.
The relevant techniques of indirect translation include transposition, modulation and
adaptation or their combination (cf. Vinay-Darbelent, 1958/1995). Given the
complexity of translation and the differences between languages, there is naturally a
plenty of scope for interference or negative transfer resulting in instances of deviation
from the collocational preferences of the target language. By analogy with a
phenomenon well-known in language teaching as faux amis we may call these
instances ‘collocational faux amis’.
Work on Czech-English combinatory dictionaries (Klégr et al., 1991, 1994,
2005) showed three principal types of collocational faux amis (CFA) depending on
which of the three main techniques of indirect translation is required instead of
compositional translation to render the SL collocation naturally. Transpositional CFA
require a grammatical shift between the SL collocation and its TL counterpart
(without change in denotative meaning), such as word-class shift (třít bídu “be
poor/destitute”: n>adj and lexical verb>copula), unit shift (phrase for word, clause for
word, etc.: srdce vynechává “heart beats irregularly”: V>V-Adv); structure shift
(change in word order, syntactic function, etc. – inspect a school “provest inspekci ve
škole”: v > nv; n> PrP; O > Adv), etc. Modulative CFA necessitate lexical variation:
one or more components of the SL collocation have to be translated by a word which
is not their usual, dictionary equivalent (such as a synonym, (co-)hyponym or
meronym), cf. protancovat boty (= shoes) “dance the soles off”; vraštit čelo (=
forehead) “knit (one’s) brows, frown”. Adaptational CFA are instances where – in the
absence of “structural and conceptual parallels between SL and TL” and “when the
situation referred to in ST does not exist in the target culture, or does not have the
same relevance or connotations” (Shuttleworth-Cowie, 1997) – the translation of a
collocation is possible only by adaptation. They are not easy to find and, moreover,
the distinction between modulation and adaptation is often very difficult to draw:
reconstitute dried milk “rozmíchat/rozpustit (sušené) mléko ve vodě”. It was found,
however, that contrary to claims that collocations are “completely lexically
determined and thus need to be memorized” (see Heid, 1994, referring to Mel’chuk,
Polguere, 1987), at least collocations of the vb-n or adj-n type seem to be largely
semantically determined and have straightforward correlates, while unpredictable
collocations form a distinct minority.
9
5. Conclusions
Believing that contrastive study of collocations is a long-overdue task, we have tried
to suggest how comparable and parallel corpora can be made use of. The proposed
synset procedure, applicable to parallel corpora, gives a comprehensive, yet succinct
account of the collocation patterns of a node and allows comparison with an
equivalent node in another language. The procedure could presumably be simplified
by using Kilgariff’s Word Sketch. It could do away with the laborious gathering of
collocates (and reduce their number), parsing and assignment of syntactic functions
and make it immediately possible to start with the key stages: division of collocates
into synsets and the actual comparison of the situation in each language. Analysis of
collocations in a parallel corpus introduces another aspect, degrees of correspondence
between collocations across languages, and raises the issue of mismatches, or
collocational faux amis.
References
Čermák, F. (2005) Abstract noun collocations: their nature in a parallel English-Czech
corpus, in G. Barnbrook, P. Danielsson and M. Mahlberg (eds), Meaningful
Texts, pp. 143–51. London-New York: Continuum.
Heid U. (1994) On Ways Words Work Together – Topics in Lexical Combinatorics,
in Euralex 1994 Proceedings, pp. 226–57. Amsterdam: Vrije Universiteit.
Hoey, M. (2005) Lexical priming. A new theory of words and language. London /
New York: Routledge.
Kilgarriff A., Sketch Engine. Available on-line from http://www.sketchengine.co.uk/
(last modified: Adam Kilgarriff, 25 May 2007).
Klégr A., N. Hronková, Z. Hron (1991) Znáte anglická slovesa? Česko-anglický
slovník nejužívanějších spojení podstatných jmen se slovesy [Do you know
your English verbs? A Czech-English dictionary of noun-verb combinations].
Praha: SPN.
Klégr A., N. Hronková (1994) Znáte anglická přídavná jména? Česko-anglický
slovník spojení podstatných jmen s přídavnými jmény [Do you know your
English adjectives? A Czech-English dictionary of noun-adjective
combinations]. Praha: Leda.
Klégr A., P. Key, N. Hronková (2005) Česko-anglický slovník spojení: podstatné
jméno a sloveso / Czech-English combinatory dictionary: noun and verb.
Praha: Karolinum.
Klégr A., P. Šaldová (2006) Kolokační faux amis, in Čermák, F., M. Šulc (eds)
Kolokace, pp. 168–77. Praha: NLN.
McCarthy, M., F. O’Dell (2005) English Collocations in Use. Cambridge: Cambridge
University Press.
McDonald, S., D. Turcato, P. McFetridge, F. Popowich, J. Toole (2004) Collocation
Discovery for Optimal Bilingual Lexicon Development. Berlin / Heidelberg:
Springer.
Nuccorini, S. (2003) Towards an ‘ideal’ Dictionary of English Collocations, in P. van
Sterkenburg (ed.) A Practical Guide to Lexicography. Amsterdam /
Philadephia: John Benjamins.
10
Partington, A. (1998), Patterns and Meanings: Using Corpora for English Language
Research and Teaching. Amsterdam / Philadelphia: John Benjamins.
Shuttleworth M., M. Cowie (1997) Dictionary of Translation Studies. Manchester: St.
Jerome.
Sinclair, J. (1991) Corpus, Concordance, Collocation. Oxford: Oxford University
Press.
Sinclair, J., S. Jones, R. Daley (2004) English Collocation Studies: The OSTI Report.
London / New York: Continuum.
Stubbs, M. (2002) Words and Phrases. Corpus Studies of Lexical Semantics. Oxford
UK / Cambridge USA: Blackwell Publishing.
Vinay J.-P., J. Darbelnet (1958/1995) Comparative Stylistics of French and English:
A Methodology for Translation (transl. J. C. Sager, M.-J. Hamel). Amsterdam
/ Philadephia: John Benjamins.
11